xref: /linux/drivers/nvme/host/fabrics.c (revision 120a201bd2ad0bffebdd2cf62c389dbba79bbfae)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * NVMe over Fabrics common host code.
4  * Copyright (c) 2015-2016 HGST, a Western Digital Company.
5  */
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 #include <linux/init.h>
8 #include <linux/miscdevice.h>
9 #include <linux/module.h>
10 #include <linux/mutex.h>
11 #include <linux/parser.h>
12 #include <linux/seq_file.h>
13 #include "nvme.h"
14 #include "fabrics.h"
15 #include <linux/nvme-keyring.h>
16 
17 static LIST_HEAD(nvmf_transports);
18 static DECLARE_RWSEM(nvmf_transports_rwsem);
19 
20 static LIST_HEAD(nvmf_hosts);
21 static DEFINE_MUTEX(nvmf_hosts_mutex);
22 
23 static struct nvmf_host *nvmf_default_host;
24 
25 static struct nvmf_host *nvmf_host_alloc(const char *hostnqn, uuid_t *id)
26 {
27 	struct nvmf_host *host;
28 
29 	host = kmalloc(sizeof(*host), GFP_KERNEL);
30 	if (!host)
31 		return NULL;
32 
33 	kref_init(&host->ref);
34 	uuid_copy(&host->id, id);
35 	strscpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
36 
37 	return host;
38 }
39 
40 static struct nvmf_host *nvmf_host_add(const char *hostnqn, uuid_t *id)
41 {
42 	struct nvmf_host *host;
43 
44 	mutex_lock(&nvmf_hosts_mutex);
45 
46 	/*
47 	 * We have defined a host as how it is perceived by the target.
48 	 * Therefore, we don't allow different Host NQNs with the same Host ID.
49 	 * Similarly, we do not allow the usage of the same Host NQN with
50 	 * different Host IDs. This'll maintain unambiguous host identification.
51 	 */
52 	list_for_each_entry(host, &nvmf_hosts, list) {
53 		bool same_hostnqn = !strcmp(host->nqn, hostnqn);
54 		bool same_hostid = uuid_equal(&host->id, id);
55 
56 		if (same_hostnqn && same_hostid) {
57 			kref_get(&host->ref);
58 			goto out_unlock;
59 		}
60 		if (same_hostnqn) {
61 			pr_err("found same hostnqn %s but different hostid %pUb\n",
62 			       hostnqn, id);
63 			host = ERR_PTR(-EINVAL);
64 			goto out_unlock;
65 		}
66 		if (same_hostid) {
67 			pr_err("found same hostid %pUb but different hostnqn %s\n",
68 			       id, hostnqn);
69 			host = ERR_PTR(-EINVAL);
70 			goto out_unlock;
71 		}
72 	}
73 
74 	host = nvmf_host_alloc(hostnqn, id);
75 	if (!host) {
76 		host = ERR_PTR(-ENOMEM);
77 		goto out_unlock;
78 	}
79 
80 	list_add_tail(&host->list, &nvmf_hosts);
81 out_unlock:
82 	mutex_unlock(&nvmf_hosts_mutex);
83 	return host;
84 }
85 
86 static struct nvmf_host *nvmf_host_default(void)
87 {
88 	struct nvmf_host *host;
89 	char nqn[NVMF_NQN_SIZE];
90 	uuid_t id;
91 
92 	uuid_gen(&id);
93 	snprintf(nqn, NVMF_NQN_SIZE,
94 		"nqn.2014-08.org.nvmexpress:uuid:%pUb", &id);
95 
96 	host = nvmf_host_alloc(nqn, &id);
97 	if (!host)
98 		return NULL;
99 
100 	mutex_lock(&nvmf_hosts_mutex);
101 	list_add_tail(&host->list, &nvmf_hosts);
102 	mutex_unlock(&nvmf_hosts_mutex);
103 
104 	return host;
105 }
106 
107 static void nvmf_host_destroy(struct kref *ref)
108 {
109 	struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
110 
111 	mutex_lock(&nvmf_hosts_mutex);
112 	list_del(&host->list);
113 	mutex_unlock(&nvmf_hosts_mutex);
114 
115 	kfree(host);
116 }
117 
118 static void nvmf_host_put(struct nvmf_host *host)
119 {
120 	if (host)
121 		kref_put(&host->ref, nvmf_host_destroy);
122 }
123 
124 /**
125  * nvmf_get_address() -  Get address/port
126  * @ctrl:	Host NVMe controller instance which we got the address
127  * @buf:	OUTPUT parameter that will contain the address/port
128  * @size:	buffer size
129  */
130 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
131 {
132 	int len = 0;
133 
134 	if (ctrl->opts->mask & NVMF_OPT_TRADDR)
135 		len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
136 	if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
137 		len += scnprintf(buf + len, size - len, "%strsvcid=%s",
138 				(len) ? "," : "", ctrl->opts->trsvcid);
139 	if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
140 		len += scnprintf(buf + len, size - len, "%shost_traddr=%s",
141 				(len) ? "," : "", ctrl->opts->host_traddr);
142 	if (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)
143 		len += scnprintf(buf + len, size - len, "%shost_iface=%s",
144 				(len) ? "," : "", ctrl->opts->host_iface);
145 	len += scnprintf(buf + len, size - len, "\n");
146 
147 	return len;
148 }
149 EXPORT_SYMBOL_GPL(nvmf_get_address);
150 
151 /**
152  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
153  * @ctrl:	Host NVMe controller instance maintaining the admin
154  *		queue used to submit the property read command to
155  *		the allocated NVMe controller resource on the target system.
156  * @off:	Starting offset value of the targeted property
157  *		register (see the fabrics section of the NVMe standard).
158  * @val:	OUTPUT parameter that will contain the value of
159  *		the property after a successful read.
160  *
161  * Used by the host system to retrieve a 32-bit capsule property value
162  * from an NVMe controller on the target system.
163  *
164  * ("Capsule property" is an "PCIe register concept" applied to the
165  * NVMe fabrics space.)
166  *
167  * Return:
168  *	0: successful read
169  *	> 0: NVMe error status code
170  *	< 0: Linux errno error code
171  */
172 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
173 {
174 	struct nvme_command cmd = { };
175 	union nvme_result res;
176 	int ret;
177 
178 	cmd.prop_get.opcode = nvme_fabrics_command;
179 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
180 	cmd.prop_get.offset = cpu_to_le32(off);
181 
182 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0,
183 			NVME_QID_ANY, 0, 0);
184 
185 	if (ret >= 0)
186 		*val = le64_to_cpu(res.u64);
187 	if (unlikely(ret != 0))
188 		dev_err(ctrl->device,
189 			"Property Get error: %d, offset %#x\n",
190 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
191 
192 	return ret;
193 }
194 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
195 
196 /**
197  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
198  * @ctrl:	Host NVMe controller instance maintaining the admin
199  *		queue used to submit the property read command to
200  *		the allocated controller resource on the target system.
201  * @off:	Starting offset value of the targeted property
202  *		register (see the fabrics section of the NVMe standard).
203  * @val:	OUTPUT parameter that will contain the value of
204  *		the property after a successful read.
205  *
206  * Used by the host system to retrieve a 64-bit capsule property value
207  * from an NVMe controller on the target system.
208  *
209  * ("Capsule property" is an "PCIe register concept" applied to the
210  * NVMe fabrics space.)
211  *
212  * Return:
213  *	0: successful read
214  *	> 0: NVMe error status code
215  *	< 0: Linux errno error code
216  */
217 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
218 {
219 	struct nvme_command cmd = { };
220 	union nvme_result res;
221 	int ret;
222 
223 	cmd.prop_get.opcode = nvme_fabrics_command;
224 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
225 	cmd.prop_get.attrib = 1;
226 	cmd.prop_get.offset = cpu_to_le32(off);
227 
228 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0,
229 			NVME_QID_ANY, 0, 0);
230 
231 	if (ret >= 0)
232 		*val = le64_to_cpu(res.u64);
233 	if (unlikely(ret != 0))
234 		dev_err(ctrl->device,
235 			"Property Get error: %d, offset %#x\n",
236 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
237 	return ret;
238 }
239 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
240 
241 /**
242  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
243  * @ctrl:	Host NVMe controller instance maintaining the admin
244  *		queue used to submit the property read command to
245  *		the allocated NVMe controller resource on the target system.
246  * @off:	Starting offset value of the targeted property
247  *		register (see the fabrics section of the NVMe standard).
248  * @val:	Input parameter that contains the value to be
249  *		written to the property.
250  *
251  * Used by the NVMe host system to write a 32-bit capsule property value
252  * to an NVMe controller on the target system.
253  *
254  * ("Capsule property" is an "PCIe register concept" applied to the
255  * NVMe fabrics space.)
256  *
257  * Return:
258  *	0: successful write
259  *	> 0: NVMe error status code
260  *	< 0: Linux errno error code
261  */
262 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
263 {
264 	struct nvme_command cmd = { };
265 	int ret;
266 
267 	cmd.prop_set.opcode = nvme_fabrics_command;
268 	cmd.prop_set.fctype = nvme_fabrics_type_property_set;
269 	cmd.prop_set.attrib = 0;
270 	cmd.prop_set.offset = cpu_to_le32(off);
271 	cmd.prop_set.value = cpu_to_le64(val);
272 
273 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0,
274 			NVME_QID_ANY, 0, 0);
275 	if (unlikely(ret))
276 		dev_err(ctrl->device,
277 			"Property Set error: %d, offset %#x\n",
278 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
279 	return ret;
280 }
281 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
282 
283 /**
284  * nvmf_log_connect_error() - Error-parsing-diagnostic print out function for
285  * 				connect() errors.
286  * @ctrl:	The specific /dev/nvmeX device that had the error.
287  * @errval:	Error code to be decoded in a more human-friendly
288  * 		printout.
289  * @offset:	For use with the NVMe error code
290  * 		NVME_SC_CONNECT_INVALID_PARAM.
291  * @cmd:	This is the SQE portion of a submission capsule.
292  * @data:	This is the "Data" portion of a submission capsule.
293  */
294 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
295 		int errval, int offset, struct nvme_command *cmd,
296 		struct nvmf_connect_data *data)
297 {
298 	int err_sctype = errval & ~NVME_SC_DNR;
299 
300 	if (errval < 0) {
301 		dev_err(ctrl->device,
302 			"Connect command failed, errno: %d\n", errval);
303 		return;
304 	}
305 
306 	switch (err_sctype) {
307 	case NVME_SC_CONNECT_INVALID_PARAM:
308 		if (offset >> 16) {
309 			char *inv_data = "Connect Invalid Data Parameter";
310 
311 			switch (offset & 0xffff) {
312 			case (offsetof(struct nvmf_connect_data, cntlid)):
313 				dev_err(ctrl->device,
314 					"%s, cntlid: %d\n",
315 					inv_data, data->cntlid);
316 				break;
317 			case (offsetof(struct nvmf_connect_data, hostnqn)):
318 				dev_err(ctrl->device,
319 					"%s, hostnqn \"%s\"\n",
320 					inv_data, data->hostnqn);
321 				break;
322 			case (offsetof(struct nvmf_connect_data, subsysnqn)):
323 				dev_err(ctrl->device,
324 					"%s, subsysnqn \"%s\"\n",
325 					inv_data, data->subsysnqn);
326 				break;
327 			default:
328 				dev_err(ctrl->device,
329 					"%s, starting byte offset: %d\n",
330 				       inv_data, offset & 0xffff);
331 				break;
332 			}
333 		} else {
334 			char *inv_sqe = "Connect Invalid SQE Parameter";
335 
336 			switch (offset) {
337 			case (offsetof(struct nvmf_connect_command, qid)):
338 				dev_err(ctrl->device,
339 				       "%s, qid %d\n",
340 					inv_sqe, cmd->connect.qid);
341 				break;
342 			default:
343 				dev_err(ctrl->device,
344 					"%s, starting byte offset: %d\n",
345 					inv_sqe, offset);
346 			}
347 		}
348 		break;
349 	case NVME_SC_CONNECT_INVALID_HOST:
350 		dev_err(ctrl->device,
351 			"Connect for subsystem %s is not allowed, hostnqn: %s\n",
352 			data->subsysnqn, data->hostnqn);
353 		break;
354 	case NVME_SC_CONNECT_CTRL_BUSY:
355 		dev_err(ctrl->device,
356 			"Connect command failed: controller is busy or not available\n");
357 		break;
358 	case NVME_SC_CONNECT_FORMAT:
359 		dev_err(ctrl->device,
360 			"Connect incompatible format: %d",
361 			cmd->connect.recfmt);
362 		break;
363 	case NVME_SC_HOST_PATH_ERROR:
364 		dev_err(ctrl->device,
365 			"Connect command failed: host path error\n");
366 		break;
367 	case NVME_SC_AUTH_REQUIRED:
368 		dev_err(ctrl->device,
369 			"Connect command failed: authentication required\n");
370 		break;
371 	default:
372 		dev_err(ctrl->device,
373 			"Connect command failed, error wo/DNR bit: %d\n",
374 			err_sctype);
375 		break;
376 	}
377 }
378 
379 static struct nvmf_connect_data *nvmf_connect_data_prep(struct nvme_ctrl *ctrl,
380 		u16 cntlid)
381 {
382 	struct nvmf_connect_data *data;
383 
384 	data = kzalloc(sizeof(*data), GFP_KERNEL);
385 	if (!data)
386 		return NULL;
387 
388 	uuid_copy(&data->hostid, &ctrl->opts->host->id);
389 	data->cntlid = cpu_to_le16(cntlid);
390 	strscpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
391 	strscpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
392 
393 	return data;
394 }
395 
396 static void nvmf_connect_cmd_prep(struct nvme_ctrl *ctrl, u16 qid,
397 		struct nvme_command *cmd)
398 {
399 	cmd->connect.opcode = nvme_fabrics_command;
400 	cmd->connect.fctype = nvme_fabrics_type_connect;
401 	cmd->connect.qid = cpu_to_le16(qid);
402 
403 	if (qid) {
404 		cmd->connect.sqsize = cpu_to_le16(ctrl->sqsize);
405 	} else {
406 		cmd->connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
407 
408 		/*
409 		 * set keep-alive timeout in seconds granularity (ms * 1000)
410 		 */
411 		cmd->connect.kato = cpu_to_le32(ctrl->kato * 1000);
412 	}
413 
414 	if (ctrl->opts->disable_sqflow)
415 		cmd->connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
416 }
417 
418 /**
419  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
420  *				API function.
421  * @ctrl:	Host nvme controller instance used to request
422  *              a new NVMe controller allocation on the target
423  *              system and  establish an NVMe Admin connection to
424  *              that controller.
425  *
426  * This function enables an NVMe host device to request a new allocation of
427  * an NVMe controller resource on a target system as well establish a
428  * fabrics-protocol connection of the NVMe Admin queue between the
429  * host system device and the allocated NVMe controller on the
430  * target system via a NVMe Fabrics "Connect" command.
431  *
432  * Return:
433  *	0: success
434  *	> 0: NVMe error status code
435  *	< 0: Linux errno error code
436  *
437  */
438 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
439 {
440 	struct nvme_command cmd = { };
441 	union nvme_result res;
442 	struct nvmf_connect_data *data;
443 	int ret;
444 	u32 result;
445 
446 	nvmf_connect_cmd_prep(ctrl, 0, &cmd);
447 
448 	data = nvmf_connect_data_prep(ctrl, 0xffff);
449 	if (!data)
450 		return -ENOMEM;
451 
452 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
453 			data, sizeof(*data), NVME_QID_ANY, 1,
454 			BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
455 	if (ret) {
456 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
457 				       &cmd, data);
458 		goto out_free_data;
459 	}
460 
461 	result = le32_to_cpu(res.u32);
462 	ctrl->cntlid = result & 0xFFFF;
463 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
464 		/* Secure concatenation is not implemented */
465 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
466 			dev_warn(ctrl->device,
467 				 "qid 0: secure concatenation is not supported\n");
468 			ret = NVME_SC_AUTH_REQUIRED;
469 			goto out_free_data;
470 		}
471 		/* Authentication required */
472 		ret = nvme_auth_negotiate(ctrl, 0);
473 		if (ret) {
474 			dev_warn(ctrl->device,
475 				 "qid 0: authentication setup failed\n");
476 			ret = NVME_SC_AUTH_REQUIRED;
477 			goto out_free_data;
478 		}
479 		ret = nvme_auth_wait(ctrl, 0);
480 		if (ret)
481 			dev_warn(ctrl->device,
482 				 "qid 0: authentication failed\n");
483 		else
484 			dev_info(ctrl->device,
485 				 "qid 0: authenticated\n");
486 	}
487 out_free_data:
488 	kfree(data);
489 	return ret;
490 }
491 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
492 
493 /**
494  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
495  *			     API function.
496  * @ctrl:	Host nvme controller instance used to establish an
497  *		NVMe I/O queue connection to the already allocated NVMe
498  *		controller on the target system.
499  * @qid:	NVMe I/O queue number for the new I/O connection between
500  *		host and target (note qid == 0 is illegal as this is
501  *		the Admin queue, per NVMe standard).
502  *
503  * This function issues a fabrics-protocol connection
504  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
505  * between the host system device and the allocated NVMe controller
506  * on the target system.
507  *
508  * Return:
509  *	0: success
510  *	> 0: NVMe error status code
511  *	< 0: Linux errno error code
512  */
513 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
514 {
515 	struct nvme_command cmd = { };
516 	struct nvmf_connect_data *data;
517 	union nvme_result res;
518 	int ret;
519 	u32 result;
520 
521 	nvmf_connect_cmd_prep(ctrl, qid, &cmd);
522 
523 	data = nvmf_connect_data_prep(ctrl, ctrl->cntlid);
524 	if (!data)
525 		return -ENOMEM;
526 
527 	ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
528 			data, sizeof(*data), qid, 1,
529 			BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
530 	if (ret) {
531 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
532 				       &cmd, data);
533 	}
534 	result = le32_to_cpu(res.u32);
535 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
536 		/* Secure concatenation is not implemented */
537 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
538 			dev_warn(ctrl->device,
539 				 "qid 0: secure concatenation is not supported\n");
540 			ret = NVME_SC_AUTH_REQUIRED;
541 			goto out_free_data;
542 		}
543 		/* Authentication required */
544 		ret = nvme_auth_negotiate(ctrl, qid);
545 		if (ret) {
546 			dev_warn(ctrl->device,
547 				 "qid %d: authentication setup failed\n", qid);
548 			ret = NVME_SC_AUTH_REQUIRED;
549 		} else {
550 			ret = nvme_auth_wait(ctrl, qid);
551 			if (ret)
552 				dev_warn(ctrl->device,
553 					 "qid %u: authentication failed\n", qid);
554 		}
555 	}
556 out_free_data:
557 	kfree(data);
558 	return ret;
559 }
560 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
561 
562 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
563 {
564 	if (ctrl->opts->max_reconnects == -1 ||
565 	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
566 		return true;
567 
568 	return false;
569 }
570 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
571 
572 /**
573  * nvmf_register_transport() - NVMe Fabrics Library registration function.
574  * @ops:	Transport ops instance to be registered to the
575  *		common fabrics library.
576  *
577  * API function that registers the type of specific transport fabric
578  * being implemented to the common NVMe fabrics library. Part of
579  * the overall init sequence of starting up a fabrics driver.
580  */
581 int nvmf_register_transport(struct nvmf_transport_ops *ops)
582 {
583 	if (!ops->create_ctrl)
584 		return -EINVAL;
585 
586 	down_write(&nvmf_transports_rwsem);
587 	list_add_tail(&ops->entry, &nvmf_transports);
588 	up_write(&nvmf_transports_rwsem);
589 
590 	return 0;
591 }
592 EXPORT_SYMBOL_GPL(nvmf_register_transport);
593 
594 /**
595  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
596  * @ops:	Transport ops instance to be unregistered from the
597  *		common fabrics library.
598  *
599  * Fabrics API function that unregisters the type of specific transport
600  * fabric being implemented from the common NVMe fabrics library.
601  * Part of the overall exit sequence of unloading the implemented driver.
602  */
603 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
604 {
605 	down_write(&nvmf_transports_rwsem);
606 	list_del(&ops->entry);
607 	up_write(&nvmf_transports_rwsem);
608 }
609 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
610 
611 static struct nvmf_transport_ops *nvmf_lookup_transport(
612 		struct nvmf_ctrl_options *opts)
613 {
614 	struct nvmf_transport_ops *ops;
615 
616 	lockdep_assert_held(&nvmf_transports_rwsem);
617 
618 	list_for_each_entry(ops, &nvmf_transports, entry) {
619 		if (strcmp(ops->name, opts->transport) == 0)
620 			return ops;
621 	}
622 
623 	return NULL;
624 }
625 
626 static struct key *nvmf_parse_key(int key_id)
627 {
628 	struct key *key;
629 
630 	if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
631 		pr_err("TLS is not supported\n");
632 		return ERR_PTR(-EINVAL);
633 	}
634 
635 	key = key_lookup(key_id);
636 	if (!IS_ERR(key))
637 		pr_err("key id %08x not found\n", key_id);
638 	else
639 		pr_debug("Using key id %08x\n", key_id);
640 	return key;
641 }
642 
643 static const match_table_t opt_tokens = {
644 	{ NVMF_OPT_TRANSPORT,		"transport=%s"		},
645 	{ NVMF_OPT_TRADDR,		"traddr=%s"		},
646 	{ NVMF_OPT_TRSVCID,		"trsvcid=%s"		},
647 	{ NVMF_OPT_NQN,			"nqn=%s"		},
648 	{ NVMF_OPT_QUEUE_SIZE,		"queue_size=%d"		},
649 	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
650 	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
651 	{ NVMF_OPT_CTRL_LOSS_TMO,	"ctrl_loss_tmo=%d"	},
652 	{ NVMF_OPT_KATO,		"keep_alive_tmo=%d"	},
653 	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
654 	{ NVMF_OPT_HOST_TRADDR,		"host_traddr=%s"	},
655 	{ NVMF_OPT_HOST_IFACE,		"host_iface=%s"		},
656 	{ NVMF_OPT_HOST_ID,		"hostid=%s"		},
657 	{ NVMF_OPT_DUP_CONNECT,		"duplicate_connect"	},
658 	{ NVMF_OPT_DISABLE_SQFLOW,	"disable_sqflow"	},
659 	{ NVMF_OPT_HDR_DIGEST,		"hdr_digest"		},
660 	{ NVMF_OPT_DATA_DIGEST,		"data_digest"		},
661 	{ NVMF_OPT_NR_WRITE_QUEUES,	"nr_write_queues=%d"	},
662 	{ NVMF_OPT_NR_POLL_QUEUES,	"nr_poll_queues=%d"	},
663 	{ NVMF_OPT_TOS,			"tos=%d"		},
664 #ifdef CONFIG_NVME_TCP_TLS
665 	{ NVMF_OPT_KEYRING,		"keyring=%d"		},
666 	{ NVMF_OPT_TLS_KEY,		"tls_key=%d"		},
667 #endif
668 	{ NVMF_OPT_FAIL_FAST_TMO,	"fast_io_fail_tmo=%d"	},
669 	{ NVMF_OPT_DISCOVERY,		"discovery"		},
670 #ifdef CONFIG_NVME_HOST_AUTH
671 	{ NVMF_OPT_DHCHAP_SECRET,	"dhchap_secret=%s"	},
672 	{ NVMF_OPT_DHCHAP_CTRL_SECRET,	"dhchap_ctrl_secret=%s"	},
673 #endif
674 #ifdef CONFIG_NVME_TCP_TLS
675 	{ NVMF_OPT_TLS,			"tls"			},
676 #endif
677 	{ NVMF_OPT_ERR,			NULL			}
678 };
679 
680 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
681 		const char *buf)
682 {
683 	substring_t args[MAX_OPT_ARGS];
684 	char *options, *o, *p;
685 	int token, ret = 0;
686 	size_t nqnlen  = 0;
687 	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO, key_id;
688 	uuid_t hostid;
689 	char hostnqn[NVMF_NQN_SIZE];
690 	struct key *key;
691 
692 	/* Set defaults */
693 	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
694 	opts->nr_io_queues = num_online_cpus();
695 	opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
696 	opts->kato = 0;
697 	opts->duplicate_connect = false;
698 	opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
699 	opts->hdr_digest = false;
700 	opts->data_digest = false;
701 	opts->tos = -1; /* < 0 == use transport default */
702 	opts->tls = false;
703 	opts->tls_key = NULL;
704 	opts->keyring = NULL;
705 
706 	options = o = kstrdup(buf, GFP_KERNEL);
707 	if (!options)
708 		return -ENOMEM;
709 
710 	/* use default host if not given by user space */
711 	uuid_copy(&hostid, &nvmf_default_host->id);
712 	strscpy(hostnqn, nvmf_default_host->nqn, NVMF_NQN_SIZE);
713 
714 	while ((p = strsep(&o, ",\n")) != NULL) {
715 		if (!*p)
716 			continue;
717 
718 		token = match_token(p, opt_tokens, args);
719 		opts->mask |= token;
720 		switch (token) {
721 		case NVMF_OPT_TRANSPORT:
722 			p = match_strdup(args);
723 			if (!p) {
724 				ret = -ENOMEM;
725 				goto out;
726 			}
727 			kfree(opts->transport);
728 			opts->transport = p;
729 			break;
730 		case NVMF_OPT_NQN:
731 			p = match_strdup(args);
732 			if (!p) {
733 				ret = -ENOMEM;
734 				goto out;
735 			}
736 			kfree(opts->subsysnqn);
737 			opts->subsysnqn = p;
738 			nqnlen = strlen(opts->subsysnqn);
739 			if (nqnlen >= NVMF_NQN_SIZE) {
740 				pr_err("%s needs to be < %d bytes\n",
741 					opts->subsysnqn, NVMF_NQN_SIZE);
742 				ret = -EINVAL;
743 				goto out;
744 			}
745 			opts->discovery_nqn =
746 				!(strcmp(opts->subsysnqn,
747 					 NVME_DISC_SUBSYS_NAME));
748 			break;
749 		case NVMF_OPT_TRADDR:
750 			p = match_strdup(args);
751 			if (!p) {
752 				ret = -ENOMEM;
753 				goto out;
754 			}
755 			kfree(opts->traddr);
756 			opts->traddr = p;
757 			break;
758 		case NVMF_OPT_TRSVCID:
759 			p = match_strdup(args);
760 			if (!p) {
761 				ret = -ENOMEM;
762 				goto out;
763 			}
764 			kfree(opts->trsvcid);
765 			opts->trsvcid = p;
766 			break;
767 		case NVMF_OPT_QUEUE_SIZE:
768 			if (match_int(args, &token)) {
769 				ret = -EINVAL;
770 				goto out;
771 			}
772 			if (token < NVMF_MIN_QUEUE_SIZE ||
773 			    token > NVMF_MAX_QUEUE_SIZE) {
774 				pr_err("Invalid queue_size %d\n", token);
775 				ret = -EINVAL;
776 				goto out;
777 			}
778 			opts->queue_size = token;
779 			break;
780 		case NVMF_OPT_NR_IO_QUEUES:
781 			if (match_int(args, &token)) {
782 				ret = -EINVAL;
783 				goto out;
784 			}
785 			if (token <= 0) {
786 				pr_err("Invalid number of IOQs %d\n", token);
787 				ret = -EINVAL;
788 				goto out;
789 			}
790 			if (opts->discovery_nqn) {
791 				pr_debug("Ignoring nr_io_queues value for discovery controller\n");
792 				break;
793 			}
794 
795 			opts->nr_io_queues = min_t(unsigned int,
796 					num_online_cpus(), token);
797 			break;
798 		case NVMF_OPT_KATO:
799 			if (match_int(args, &token)) {
800 				ret = -EINVAL;
801 				goto out;
802 			}
803 
804 			if (token < 0) {
805 				pr_err("Invalid keep_alive_tmo %d\n", token);
806 				ret = -EINVAL;
807 				goto out;
808 			} else if (token == 0 && !opts->discovery_nqn) {
809 				/* Allowed for debug */
810 				pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
811 			}
812 			opts->kato = token;
813 			break;
814 		case NVMF_OPT_CTRL_LOSS_TMO:
815 			if (match_int(args, &token)) {
816 				ret = -EINVAL;
817 				goto out;
818 			}
819 
820 			if (token < 0)
821 				pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
822 			ctrl_loss_tmo = token;
823 			break;
824 		case NVMF_OPT_FAIL_FAST_TMO:
825 			if (match_int(args, &token)) {
826 				ret = -EINVAL;
827 				goto out;
828 			}
829 
830 			if (token >= 0)
831 				pr_warn("I/O fail on reconnect controller after %d sec\n",
832 					token);
833 			else
834 				token = -1;
835 
836 			opts->fast_io_fail_tmo = token;
837 			break;
838 		case NVMF_OPT_HOSTNQN:
839 			if (opts->host) {
840 				pr_err("hostnqn already user-assigned: %s\n",
841 				       opts->host->nqn);
842 				ret = -EADDRINUSE;
843 				goto out;
844 			}
845 			p = match_strdup(args);
846 			if (!p) {
847 				ret = -ENOMEM;
848 				goto out;
849 			}
850 			nqnlen = strlen(p);
851 			if (nqnlen >= NVMF_NQN_SIZE) {
852 				pr_err("%s needs to be < %d bytes\n",
853 					p, NVMF_NQN_SIZE);
854 				kfree(p);
855 				ret = -EINVAL;
856 				goto out;
857 			}
858 			strscpy(hostnqn, p, NVMF_NQN_SIZE);
859 			kfree(p);
860 			break;
861 		case NVMF_OPT_RECONNECT_DELAY:
862 			if (match_int(args, &token)) {
863 				ret = -EINVAL;
864 				goto out;
865 			}
866 			if (token <= 0) {
867 				pr_err("Invalid reconnect_delay %d\n", token);
868 				ret = -EINVAL;
869 				goto out;
870 			}
871 			opts->reconnect_delay = token;
872 			break;
873 		case NVMF_OPT_HOST_TRADDR:
874 			p = match_strdup(args);
875 			if (!p) {
876 				ret = -ENOMEM;
877 				goto out;
878 			}
879 			kfree(opts->host_traddr);
880 			opts->host_traddr = p;
881 			break;
882 		case NVMF_OPT_HOST_IFACE:
883 			p = match_strdup(args);
884 			if (!p) {
885 				ret = -ENOMEM;
886 				goto out;
887 			}
888 			kfree(opts->host_iface);
889 			opts->host_iface = p;
890 			break;
891 		case NVMF_OPT_HOST_ID:
892 			p = match_strdup(args);
893 			if (!p) {
894 				ret = -ENOMEM;
895 				goto out;
896 			}
897 			ret = uuid_parse(p, &hostid);
898 			if (ret) {
899 				pr_err("Invalid hostid %s\n", p);
900 				ret = -EINVAL;
901 				kfree(p);
902 				goto out;
903 			}
904 			kfree(p);
905 			break;
906 		case NVMF_OPT_DUP_CONNECT:
907 			opts->duplicate_connect = true;
908 			break;
909 		case NVMF_OPT_DISABLE_SQFLOW:
910 			opts->disable_sqflow = true;
911 			break;
912 		case NVMF_OPT_HDR_DIGEST:
913 			opts->hdr_digest = true;
914 			break;
915 		case NVMF_OPT_DATA_DIGEST:
916 			opts->data_digest = true;
917 			break;
918 		case NVMF_OPT_NR_WRITE_QUEUES:
919 			if (match_int(args, &token)) {
920 				ret = -EINVAL;
921 				goto out;
922 			}
923 			if (token <= 0) {
924 				pr_err("Invalid nr_write_queues %d\n", token);
925 				ret = -EINVAL;
926 				goto out;
927 			}
928 			opts->nr_write_queues = token;
929 			break;
930 		case NVMF_OPT_NR_POLL_QUEUES:
931 			if (match_int(args, &token)) {
932 				ret = -EINVAL;
933 				goto out;
934 			}
935 			if (token <= 0) {
936 				pr_err("Invalid nr_poll_queues %d\n", token);
937 				ret = -EINVAL;
938 				goto out;
939 			}
940 			opts->nr_poll_queues = token;
941 			break;
942 		case NVMF_OPT_TOS:
943 			if (match_int(args, &token)) {
944 				ret = -EINVAL;
945 				goto out;
946 			}
947 			if (token < 0) {
948 				pr_err("Invalid type of service %d\n", token);
949 				ret = -EINVAL;
950 				goto out;
951 			}
952 			if (token > 255) {
953 				pr_warn("Clamping type of service to 255\n");
954 				token = 255;
955 			}
956 			opts->tos = token;
957 			break;
958 		case NVMF_OPT_KEYRING:
959 			if (match_int(args, &key_id) || key_id <= 0) {
960 				ret = -EINVAL;
961 				goto out;
962 			}
963 			key = nvmf_parse_key(key_id);
964 			if (IS_ERR(key)) {
965 				ret = PTR_ERR(key);
966 				goto out;
967 			}
968 			key_put(opts->keyring);
969 			opts->keyring = key;
970 			break;
971 		case NVMF_OPT_TLS_KEY:
972 			if (match_int(args, &key_id) || key_id <= 0) {
973 				ret = -EINVAL;
974 				goto out;
975 			}
976 			key = nvmf_parse_key(key_id);
977 			if (IS_ERR(key)) {
978 				ret = PTR_ERR(key);
979 				goto out;
980 			}
981 			key_put(opts->tls_key);
982 			opts->tls_key = key;
983 			break;
984 		case NVMF_OPT_DISCOVERY:
985 			opts->discovery_nqn = true;
986 			break;
987 		case NVMF_OPT_DHCHAP_SECRET:
988 			p = match_strdup(args);
989 			if (!p) {
990 				ret = -ENOMEM;
991 				goto out;
992 			}
993 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
994 				pr_err("Invalid DH-CHAP secret %s\n", p);
995 				ret = -EINVAL;
996 				goto out;
997 			}
998 			kfree(opts->dhchap_secret);
999 			opts->dhchap_secret = p;
1000 			break;
1001 		case NVMF_OPT_DHCHAP_CTRL_SECRET:
1002 			p = match_strdup(args);
1003 			if (!p) {
1004 				ret = -ENOMEM;
1005 				goto out;
1006 			}
1007 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
1008 				pr_err("Invalid DH-CHAP secret %s\n", p);
1009 				ret = -EINVAL;
1010 				goto out;
1011 			}
1012 			kfree(opts->dhchap_ctrl_secret);
1013 			opts->dhchap_ctrl_secret = p;
1014 			break;
1015 		case NVMF_OPT_TLS:
1016 			if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
1017 				pr_err("TLS is not supported\n");
1018 				ret = -EINVAL;
1019 				goto out;
1020 			}
1021 			opts->tls = true;
1022 			break;
1023 		default:
1024 			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
1025 				p);
1026 			ret = -EINVAL;
1027 			goto out;
1028 		}
1029 	}
1030 
1031 	if (opts->discovery_nqn) {
1032 		opts->nr_io_queues = 0;
1033 		opts->nr_write_queues = 0;
1034 		opts->nr_poll_queues = 0;
1035 		opts->duplicate_connect = true;
1036 	} else {
1037 		if (!opts->kato)
1038 			opts->kato = NVME_DEFAULT_KATO;
1039 	}
1040 	if (ctrl_loss_tmo < 0) {
1041 		opts->max_reconnects = -1;
1042 	} else {
1043 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
1044 						opts->reconnect_delay);
1045 		if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
1046 			pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
1047 				opts->fast_io_fail_tmo, ctrl_loss_tmo);
1048 	}
1049 
1050 	opts->host = nvmf_host_add(hostnqn, &hostid);
1051 	if (IS_ERR(opts->host)) {
1052 		ret = PTR_ERR(opts->host);
1053 		opts->host = NULL;
1054 		goto out;
1055 	}
1056 
1057 out:
1058 	kfree(options);
1059 	return ret;
1060 }
1061 
1062 void nvmf_set_io_queues(struct nvmf_ctrl_options *opts, u32 nr_io_queues,
1063 			u32 io_queues[HCTX_MAX_TYPES])
1064 {
1065 	if (opts->nr_write_queues && opts->nr_io_queues < nr_io_queues) {
1066 		/*
1067 		 * separate read/write queues
1068 		 * hand out dedicated default queues only after we have
1069 		 * sufficient read queues.
1070 		 */
1071 		io_queues[HCTX_TYPE_READ] = opts->nr_io_queues;
1072 		nr_io_queues -= io_queues[HCTX_TYPE_READ];
1073 		io_queues[HCTX_TYPE_DEFAULT] =
1074 			min(opts->nr_write_queues, nr_io_queues);
1075 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1076 	} else {
1077 		/*
1078 		 * shared read/write queues
1079 		 * either no write queues were requested, or we don't have
1080 		 * sufficient queue count to have dedicated default queues.
1081 		 */
1082 		io_queues[HCTX_TYPE_DEFAULT] =
1083 			min(opts->nr_io_queues, nr_io_queues);
1084 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1085 	}
1086 
1087 	if (opts->nr_poll_queues && nr_io_queues) {
1088 		/* map dedicated poll queues only if we have queues left */
1089 		io_queues[HCTX_TYPE_POLL] =
1090 			min(opts->nr_poll_queues, nr_io_queues);
1091 	}
1092 }
1093 EXPORT_SYMBOL_GPL(nvmf_set_io_queues);
1094 
1095 void nvmf_map_queues(struct blk_mq_tag_set *set, struct nvme_ctrl *ctrl,
1096 		     u32 io_queues[HCTX_MAX_TYPES])
1097 {
1098 	struct nvmf_ctrl_options *opts = ctrl->opts;
1099 
1100 	if (opts->nr_write_queues && io_queues[HCTX_TYPE_READ]) {
1101 		/* separate read/write queues */
1102 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1103 			io_queues[HCTX_TYPE_DEFAULT];
1104 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1105 		set->map[HCTX_TYPE_READ].nr_queues =
1106 			io_queues[HCTX_TYPE_READ];
1107 		set->map[HCTX_TYPE_READ].queue_offset =
1108 			io_queues[HCTX_TYPE_DEFAULT];
1109 	} else {
1110 		/* shared read/write queues */
1111 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1112 			io_queues[HCTX_TYPE_DEFAULT];
1113 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1114 		set->map[HCTX_TYPE_READ].nr_queues =
1115 			io_queues[HCTX_TYPE_DEFAULT];
1116 		set->map[HCTX_TYPE_READ].queue_offset = 0;
1117 	}
1118 
1119 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1120 	blk_mq_map_queues(&set->map[HCTX_TYPE_READ]);
1121 	if (opts->nr_poll_queues && io_queues[HCTX_TYPE_POLL]) {
1122 		/* map dedicated poll queues only if we have queues left */
1123 		set->map[HCTX_TYPE_POLL].nr_queues = io_queues[HCTX_TYPE_POLL];
1124 		set->map[HCTX_TYPE_POLL].queue_offset =
1125 			io_queues[HCTX_TYPE_DEFAULT] +
1126 			io_queues[HCTX_TYPE_READ];
1127 		blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
1128 	}
1129 
1130 	dev_info(ctrl->device,
1131 		"mapped %d/%d/%d default/read/poll queues.\n",
1132 		io_queues[HCTX_TYPE_DEFAULT],
1133 		io_queues[HCTX_TYPE_READ],
1134 		io_queues[HCTX_TYPE_POLL]);
1135 }
1136 EXPORT_SYMBOL_GPL(nvmf_map_queues);
1137 
1138 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
1139 		unsigned int required_opts)
1140 {
1141 	if ((opts->mask & required_opts) != required_opts) {
1142 		unsigned int i;
1143 
1144 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1145 			if ((opt_tokens[i].token & required_opts) &&
1146 			    !(opt_tokens[i].token & opts->mask)) {
1147 				pr_warn("missing parameter '%s'\n",
1148 					opt_tokens[i].pattern);
1149 			}
1150 		}
1151 
1152 		return -EINVAL;
1153 	}
1154 
1155 	return 0;
1156 }
1157 
1158 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
1159 		struct nvmf_ctrl_options *opts)
1160 {
1161 	if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
1162 	    strcmp(opts->traddr, ctrl->opts->traddr) ||
1163 	    strcmp(opts->trsvcid, ctrl->opts->trsvcid))
1164 		return false;
1165 
1166 	/*
1167 	 * Checking the local address or host interfaces is rough.
1168 	 *
1169 	 * In most cases, none is specified and the host port or
1170 	 * host interface is selected by the stack.
1171 	 *
1172 	 * Assume no match if:
1173 	 * -  local address or host interface is specified and address
1174 	 *    or host interface is not the same
1175 	 * -  local address or host interface is not specified but
1176 	 *    remote is, or vice versa (admin using specific
1177 	 *    host_traddr/host_iface when it matters).
1178 	 */
1179 	if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
1180 	    (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1181 		if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
1182 			return false;
1183 	} else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
1184 		   (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1185 		return false;
1186 	}
1187 
1188 	if ((opts->mask & NVMF_OPT_HOST_IFACE) &&
1189 	    (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1190 		if (strcmp(opts->host_iface, ctrl->opts->host_iface))
1191 			return false;
1192 	} else if ((opts->mask & NVMF_OPT_HOST_IFACE) ||
1193 		   (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1194 		return false;
1195 	}
1196 
1197 	return true;
1198 }
1199 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
1200 
1201 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
1202 		unsigned int allowed_opts)
1203 {
1204 	if (opts->mask & ~allowed_opts) {
1205 		unsigned int i;
1206 
1207 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1208 			if ((opt_tokens[i].token & opts->mask) &&
1209 			    (opt_tokens[i].token & ~allowed_opts)) {
1210 				pr_warn("invalid parameter '%s'\n",
1211 					opt_tokens[i].pattern);
1212 			}
1213 		}
1214 
1215 		return -EINVAL;
1216 	}
1217 
1218 	return 0;
1219 }
1220 
1221 void nvmf_free_options(struct nvmf_ctrl_options *opts)
1222 {
1223 	nvmf_host_put(opts->host);
1224 	key_put(opts->keyring);
1225 	key_put(opts->tls_key);
1226 	kfree(opts->transport);
1227 	kfree(opts->traddr);
1228 	kfree(opts->trsvcid);
1229 	kfree(opts->subsysnqn);
1230 	kfree(opts->host_traddr);
1231 	kfree(opts->host_iface);
1232 	kfree(opts->dhchap_secret);
1233 	kfree(opts->dhchap_ctrl_secret);
1234 	kfree(opts);
1235 }
1236 EXPORT_SYMBOL_GPL(nvmf_free_options);
1237 
1238 #define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
1239 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
1240 				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
1241 				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
1242 				 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\
1243 				 NVMF_OPT_FAIL_FAST_TMO | NVMF_OPT_DHCHAP_SECRET |\
1244 				 NVMF_OPT_DHCHAP_CTRL_SECRET)
1245 
1246 static struct nvme_ctrl *
1247 nvmf_create_ctrl(struct device *dev, const char *buf)
1248 {
1249 	struct nvmf_ctrl_options *opts;
1250 	struct nvmf_transport_ops *ops;
1251 	struct nvme_ctrl *ctrl;
1252 	int ret;
1253 
1254 	opts = kzalloc(sizeof(*opts), GFP_KERNEL);
1255 	if (!opts)
1256 		return ERR_PTR(-ENOMEM);
1257 
1258 	ret = nvmf_parse_options(opts, buf);
1259 	if (ret)
1260 		goto out_free_opts;
1261 
1262 
1263 	request_module("nvme-%s", opts->transport);
1264 
1265 	/*
1266 	 * Check the generic options first as we need a valid transport for
1267 	 * the lookup below.  Then clear the generic flags so that transport
1268 	 * drivers don't have to care about them.
1269 	 */
1270 	ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1271 	if (ret)
1272 		goto out_free_opts;
1273 	opts->mask &= ~NVMF_REQUIRED_OPTS;
1274 
1275 	down_read(&nvmf_transports_rwsem);
1276 	ops = nvmf_lookup_transport(opts);
1277 	if (!ops) {
1278 		pr_info("no handler found for transport %s.\n",
1279 			opts->transport);
1280 		ret = -EINVAL;
1281 		goto out_unlock;
1282 	}
1283 
1284 	if (!try_module_get(ops->module)) {
1285 		ret = -EBUSY;
1286 		goto out_unlock;
1287 	}
1288 	up_read(&nvmf_transports_rwsem);
1289 
1290 	ret = nvmf_check_required_opts(opts, ops->required_opts);
1291 	if (ret)
1292 		goto out_module_put;
1293 	ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1294 				ops->allowed_opts | ops->required_opts);
1295 	if (ret)
1296 		goto out_module_put;
1297 
1298 	ctrl = ops->create_ctrl(dev, opts);
1299 	if (IS_ERR(ctrl)) {
1300 		ret = PTR_ERR(ctrl);
1301 		goto out_module_put;
1302 	}
1303 
1304 	module_put(ops->module);
1305 	return ctrl;
1306 
1307 out_module_put:
1308 	module_put(ops->module);
1309 	goto out_free_opts;
1310 out_unlock:
1311 	up_read(&nvmf_transports_rwsem);
1312 out_free_opts:
1313 	nvmf_free_options(opts);
1314 	return ERR_PTR(ret);
1315 }
1316 
1317 static struct class *nvmf_class;
1318 static struct device *nvmf_device;
1319 static DEFINE_MUTEX(nvmf_dev_mutex);
1320 
1321 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1322 		size_t count, loff_t *pos)
1323 {
1324 	struct seq_file *seq_file = file->private_data;
1325 	struct nvme_ctrl *ctrl;
1326 	const char *buf;
1327 	int ret = 0;
1328 
1329 	if (count > PAGE_SIZE)
1330 		return -ENOMEM;
1331 
1332 	buf = memdup_user_nul(ubuf, count);
1333 	if (IS_ERR(buf))
1334 		return PTR_ERR(buf);
1335 
1336 	mutex_lock(&nvmf_dev_mutex);
1337 	if (seq_file->private) {
1338 		ret = -EINVAL;
1339 		goto out_unlock;
1340 	}
1341 
1342 	ctrl = nvmf_create_ctrl(nvmf_device, buf);
1343 	if (IS_ERR(ctrl)) {
1344 		ret = PTR_ERR(ctrl);
1345 		goto out_unlock;
1346 	}
1347 
1348 	seq_file->private = ctrl;
1349 
1350 out_unlock:
1351 	mutex_unlock(&nvmf_dev_mutex);
1352 	kfree(buf);
1353 	return ret ? ret : count;
1354 }
1355 
1356 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file)
1357 {
1358 	const struct match_token *tok;
1359 	int idx;
1360 
1361 	/*
1362 	 * Add dummy entries for instance and cntlid to
1363 	 * signal an invalid/non-existing controller
1364 	 */
1365 	seq_puts(seq_file, "instance=-1,cntlid=-1");
1366 	for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) {
1367 		tok = &opt_tokens[idx];
1368 		if (tok->token == NVMF_OPT_ERR)
1369 			continue;
1370 		seq_puts(seq_file, ",");
1371 		seq_puts(seq_file, tok->pattern);
1372 	}
1373 	seq_puts(seq_file, "\n");
1374 }
1375 
1376 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1377 {
1378 	struct nvme_ctrl *ctrl;
1379 
1380 	mutex_lock(&nvmf_dev_mutex);
1381 	ctrl = seq_file->private;
1382 	if (!ctrl) {
1383 		__nvmf_concat_opt_tokens(seq_file);
1384 		goto out_unlock;
1385 	}
1386 
1387 	seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1388 			ctrl->instance, ctrl->cntlid);
1389 
1390 out_unlock:
1391 	mutex_unlock(&nvmf_dev_mutex);
1392 	return 0;
1393 }
1394 
1395 static int nvmf_dev_open(struct inode *inode, struct file *file)
1396 {
1397 	/*
1398 	 * The miscdevice code initializes file->private_data, but doesn't
1399 	 * make use of it later.
1400 	 */
1401 	file->private_data = NULL;
1402 	return single_open(file, nvmf_dev_show, NULL);
1403 }
1404 
1405 static int nvmf_dev_release(struct inode *inode, struct file *file)
1406 {
1407 	struct seq_file *seq_file = file->private_data;
1408 	struct nvme_ctrl *ctrl = seq_file->private;
1409 
1410 	if (ctrl)
1411 		nvme_put_ctrl(ctrl);
1412 	return single_release(inode, file);
1413 }
1414 
1415 static const struct file_operations nvmf_dev_fops = {
1416 	.owner		= THIS_MODULE,
1417 	.write		= nvmf_dev_write,
1418 	.read		= seq_read,
1419 	.open		= nvmf_dev_open,
1420 	.release	= nvmf_dev_release,
1421 };
1422 
1423 static struct miscdevice nvmf_misc = {
1424 	.minor		= MISC_DYNAMIC_MINOR,
1425 	.name           = "nvme-fabrics",
1426 	.fops		= &nvmf_dev_fops,
1427 };
1428 
1429 static int __init nvmf_init(void)
1430 {
1431 	int ret;
1432 
1433 	nvmf_default_host = nvmf_host_default();
1434 	if (!nvmf_default_host)
1435 		return -ENOMEM;
1436 
1437 	nvmf_class = class_create("nvme-fabrics");
1438 	if (IS_ERR(nvmf_class)) {
1439 		pr_err("couldn't register class nvme-fabrics\n");
1440 		ret = PTR_ERR(nvmf_class);
1441 		goto out_free_host;
1442 	}
1443 
1444 	nvmf_device =
1445 		device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1446 	if (IS_ERR(nvmf_device)) {
1447 		pr_err("couldn't create nvme-fabrics device!\n");
1448 		ret = PTR_ERR(nvmf_device);
1449 		goto out_destroy_class;
1450 	}
1451 
1452 	ret = misc_register(&nvmf_misc);
1453 	if (ret) {
1454 		pr_err("couldn't register misc device: %d\n", ret);
1455 		goto out_destroy_device;
1456 	}
1457 
1458 	return 0;
1459 
1460 out_destroy_device:
1461 	device_destroy(nvmf_class, MKDEV(0, 0));
1462 out_destroy_class:
1463 	class_destroy(nvmf_class);
1464 out_free_host:
1465 	nvmf_host_put(nvmf_default_host);
1466 	return ret;
1467 }
1468 
1469 static void __exit nvmf_exit(void)
1470 {
1471 	misc_deregister(&nvmf_misc);
1472 	device_destroy(nvmf_class, MKDEV(0, 0));
1473 	class_destroy(nvmf_class);
1474 	nvmf_host_put(nvmf_default_host);
1475 
1476 	BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1477 	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1478 	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1479 	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1480 	BUILD_BUG_ON(sizeof(struct nvmf_auth_send_command) != 64);
1481 	BUILD_BUG_ON(sizeof(struct nvmf_auth_receive_command) != 64);
1482 	BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1483 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_negotiate_data) != 8);
1484 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_challenge_data) != 16);
1485 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_reply_data) != 16);
1486 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success1_data) != 16);
1487 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success2_data) != 16);
1488 }
1489 
1490 MODULE_LICENSE("GPL v2");
1491 
1492 module_init(nvmf_init);
1493 module_exit(nvmf_exit);
1494